Multiple Sclerosis(MS) is an autoimmune disease characterized by focal and diffuse brain inflammation, degeneration and repair in the central nervous system. Extended demyelination in the cerebellum has been shown ex vivo in patients with advanced disease¹, and a number of recent studies evidenced cerebellar grey-matter(GM) and global cerebellar atrophy at various disease stages^2,3. Nevertheless, while focal white-matter(WM) pathology in the cerebellum has been extensively investigated, there are only few studies focusing on focal cortical cerebellar pathology in vivo⁴. Due to the complex structure of the cerebellum, whose folia are made by an inner strip of WM bounded by a thin cortical layer, ultra-high-field MRI might be of benefit to detect cerebellar lesions both in WM and in the cortex. In this work, we studied a cohort of early-stage MS patients and evaluated the impact of focal cerebellar pathology using MP2RAGE^5,6 at ultra-high(7T) and high(3T) field MRI. Next, we assessed the relationship between cerebellar lesion number and volume, as obtained using 3T and 7T images, and clinical scores in patients.

Nineteen patients (4 males, 15 females, median age 35 years, age range: 21-46 years) with early RRMS (disease duration < 5 years from diagnosis) and mild disability (median Expand Disability Status Scale-EDSS 1.5, range:1-2) underwent MR examinations on a 3T(MAGNETOM Trio a Tim system, Siemens Healthcare, Germany) and on a research 7T system (Siemens Healthcare, Germany) using a 32-channel head coil. From the MRI acquisition protocol, we used the high-resolution Magnetization-Prepared 2 Rapid Acquisitions Gradient Echo (MP2RAGE)⁶ with the following parameters:

- 3T: TR/TI1/TI2=5000/700/2500ms and voxel size=1.0x1.0x1.2mm³;

- 7T: TR/TI1/TI2=6000/750/2350ms and voxel size=0.75x0.75x0.9mm³.

Each subject underwent a neurological examination including the following cognitive and behavioral tests: 1)Brief Repeatable Battery of Neuropsychological Tests(BRB-N), which examines verbal and spatial memory, sustained attention, information processing speed, and verbal fluency on semantic cues; 2)Multiple Sclerosis Functional Composite(MSFC) scores to quantify motor and cognitive performance, and 3)EDSS. Cerebellum segmentation was obtained using the MorphoBox^7,8 prototype. Lesion maps for each patient were obtained manually and automatically. Manual detection was performed by two experts (a neurologist and a radiologist) in the MP2RAGE image at both 3T and 7T. Automated lesion detection was performed by an in-house automated tool⁹, only for use in this research study. Total lesion load(TLL), total lesion volume(TLV) and mean lesion volume(MLV) per patient were estimated at 3T and 7T. Correlation between cognitive and behavioral tests with the TLL, TLV and MLV were performed using the Spearman test adjusted for False Discovery Rate(FDR). The performance of the automated segmentation was evaluated for both scenarios using a leave-one-out cross validation. Detection rate(DR, number of detected lesions/total manual segmented lesions) was obtained for each scenario. Finally, statistical difference was computed using the Wilcoxon signed-rank test.

Illustration of original data, manual and automated segmentation is reported in Figure 1. We observed a significantly higher TLL at 7T compared to 3T(average±SD, 7T: 9±9 lesions, range:0–31; 3T: 5±8 lesions, range:0-29,P<0.001,Figure 2). And remarkably, MPRAGE at 7T showed 33% more WM lesions and only 5% more GM lesions. The average TLV per patient at 7T was 0.29±0.54mL(range: 0-2.20 mL) and did not significantly differ from TLV at 3T(0.29±0.61 mL,range: 0-2.54 mL,P=0.231). However, average MLV in the entire patients cohort was lower at 7T(0.022±0.018 mL) compared to 3T(0.041±0.048 mL,P<0.05). No correlation was found between the lesion measures at 3T and 7T and motor performance tests. Yet interestingly, negative correlations with MLV from manual segmentations on 7T with some cognitive tests (SRTR:ρ=-0.5,P=0.0063; SRTD:ρ=-0.7,P=0.0005; VIST:ρ=-0.5,P=0.0068; and VISD:ρ=-0.5,P=0.0069 after FDR correction,Figure 3) were observed. Automatic detection rates on MP2RAGE images were similar at both field strengths (7T:median DR 72%; 3T:median DR 67%,P=0.534,Figure 4).We have previously shown that the MP2RAGE sequence is sensitive to both WM and GM pathology in MS patients⁶. Our current study demonstrates the benefits of using MP2RAGE at ultra-high field to assess the presence and clinical impact of cerebellar lesions at early disease stages. MP2RAGE at 7T evidenced a higher lesion load than MP2RAGE at 3T, probably due to improved spatial resolution, lower partial-volume effects and higher contrast-to-noise. This was significant for manual lesion segmentation but failed to reach significance for automated lesion count using an in-house software. Future efforts will be devoted to improve automated lesion detection in the cerebellum at 7T by combining MP2RAGE with other contrasts like FLAIR¹⁰. Notably, significant correlations were found between 7T cerebellar lesion volume and verbal and visual memory performances in patients, suggesting that 7T MRI may have an important clinical value in multiple sclerosis for complex brain regions such as cerebellum.